Mean Back Relaxation in Stochastic Dynamics

• Mean Back Relaxation (MBR) is a three-time statistic that evaluates the ratio between future and past displacements in stochastic processes to detect non-equilibrium biases.
• It utilizes multipoint correlations and a cutoff to regularize small fluctuations, providing a robust marker for violations of detailed balance in confined and active systems.
• MBR is applied in experimental and model-based studies to link passive microrheology with fluctuation–dissipation theorem violations, offering insights into cellular active dynamics and effective energy scales.

Mean Back Relaxation (MBR) is a three-time statistic for stochastic dynamics that quantifies how a future displacement relaxes relative to a preceding displacement. For a scalar process $x(t)$, it couples the values at past, present, and future times through a signed ratio, and its long-time behavior can serve as a marker of broken detailed balance under appropriate stationarity and ergodicity assumptions. In confined equilibrium-like settings, the long-time value is $1/2$; departures from $1/2$ indicate non-equilibrium bias, although this interpretation becomes heuristic when a well-defined mean position is absent, as in many intracellular trajectories. MBR has been developed both as a general diagnostic grounded in multipoint correlations and as a practical analysis tool for probe-particle data in living and passivated cells, where it is also connected phenomenologically to fluctuation–dissipation theorem violation via an effective energy scale (Knotz et al., 2023, Knotz et al., 8 Jul 2025).

1. Definition and basic interpretation

For a stochastic process sampled at three times $t_1=-\tau$, $t_2=0$, and $t_3=t$, with $\tau>0$ and $t>0$, define the past displacement

$d = x(0)-x(-\tau)$

and the future displacement

$b = x(t)-x(0).$

MBR is the conditional mean

$1/2$0

where the cutoff $1/2$1 excludes events with small $1/2$2, both to regularize the ratio and to introduce a controlled length scale. The conventional minus sign ensures that future motion opposite to the past displacement contributes positively.

An equivalent expression uses the three-point probability density $1/2$3: $1/2$4 with normalization

$1/2$5

This definition makes explicit that MBR is not a two-time correlator but a genuinely three-time observable. It measures whether the process tends, on average, to move back against the previous increment. If microscopic reversibility and relaxation toward a steady mean position hold, the average future displacement cancels half of the past displacement in the long-time limit. If persistent drifts, active forces, or non-reciprocal couplings are present, the ratio is biased away from that value. In the formulation of the later experimental work, excess back relaxation corresponds to values $1/2$6, whereas persistent forward bias corresponds to values $1/2$7 (Knotz et al., 8 Jul 2025).

2. Detailed balance, Gaussian structure, and asymptotic results

A central result is the long-time detailed-balance criterion. If $1/2$8 has a stationary distribution with finite, time-independent mean $1/2$9, and the process is ergodic with finite relaxation times, then detailed balance implies

$1/2$0

Under these assumptions, deviations of the long-time value from $1/2$1 mark broken detailed balance. In intracellular experiments, however, the mean position is often not well-defined; in that case the criterion remains useful as a comparative non-equilibrium marker but is no longer rigorous (Knotz et al., 8 Jul 2025).

For stationary Gaussian processes, MBR simplifies substantially. It becomes independent of the cutoff $1/2$2 and can be expressed entirely through the mean-squared displacement

$1/2$3

$1/2$4

In the small-$1/2$5 limit,

$1/2$6

where $1/2$7 is the short-time diffusion coefficient.

These Gaussian formulas make several standard examples transparent. For free Brownian motion, $1/2$8, so MBR vanishes for all $1/2$9 and $t_1=-\tau$0. For an Ornstein–Uhlenbeck process in a harmonic trap, MBR approaches $t_1=-\tau$1 monotonically at long times and is independent of $t_1=-\tau$2. More generally, for linear Gaussian equilibrium dynamics with

$t_1=-\tau$3

MBR remains between $t_1=-\tau$4 and $t_1=-\tau$5 and approaches $t_1=-\tau$6 monotonically in $t_1=-\tau$7. These examples clarify that the long-time $t_1=-\tau$8 diagnostic applies in confined settings, whereas in bulk or weak-confinement regimes MBR may differ from $t_1=-\tau$9 without implying broken detailed balance (Knotz et al., 2023).

A further asymptotic statement concerns time reversal. In the long-time limit, forward and reversed MBR obey the sum rule

$t_2=0$0

In equilibrium, where pathwise entropy production vanishes, the forward and reversed quantities coincide, and each equals $t_2=0$1 (Knotz et al., 2023).

3. Relation to multipoint densities and fluctuation–dissipation violation

The original theoretical development places MBR in a broader multipoint-correlation framework. Using a path-integral formalism with path probability $t_2=0$2 and a generating functional

$t_2=0$3

conditioning on the past displacement $t_2=0$4 via

$t_2=0$5

yields a conditioned generating functional

$t_2=0$6

This makes the conditioned mean displacement—and hence MBR—expressible through characteristic functions and, equivalently, multipoint density correlations. One consequence is that MBR can be generalized beyond positions to other stochastic observables (Knotz et al., 2023).

A particularly important extension is to microscopic densities. For an $t_2=0$7-particle system in one dimension, define

$t_2=0$8

The corresponding density MBR,

$t_2=0$9

inherits the same long-time $t_3=t$0 criterion under detailed balance and the same forward-plus-reversed sum rule. Because $t_3=t$1 remains well-defined even in bulk many-body systems, this density-based construction extends the diagnostic to settings where position-based MBR becomes problematic.

The later intracellular study connects MBR to fluctuation–dissipation theorem (FDT) violation. At equilibrium,

$t_3=t$2

where $t_3=t$3 is the power spectrum and $t_3=t$4 is the imaginary part of the response. Out of equilibrium, an effective energy is introduced through

$t_3=t$5

Empirically in cells,

$t_3=t$6

MBR provides a purely passive, time-domain, three-point statistic that correlates with the amplitude $t_3=t$7 of this FDT violation (Knotz et al., 8 Jul 2025).

4. Phenomenological and model-based connection to effective energy

Across HeLa, A549, C2C12, CT26, HoxB8, MDCK, and passivated HeLa, the amplitude $t_3=t$8 extracted from active–passive microrheology correlates linearly with the long-time MBR at fixed conditioning time: $t_3=t$9 The slopes $\tau>0$0 are negative, because $\tau>0$1 while $\tau>0$2 in living cells, and their magnitude decreases with $\tau>0$3. This linearity is reported to hold robustly over

$\tau>0$4

with excellent fits,

$\tau>0$5

(Knotz et al., 8 Jul 2025).

The same work analyzes this relation in the linear Gaussian Random Horse and Cart (RHC) model: $\tau>0$6 with thermal noises satisfying Einstein relations $\tau>0$7 and an active trap diffusion $\tau>0$8. In this model the low-frequency effective energy scales as

$\tau>0$9

At fixed $t>0$0, the long-time MBR is

$t>0$1

where $t>0$2 are the relaxation rates and

$t>0$3

Eliminating $t>0$4 in favor of $t>0$5 gives the exact relation

$t>0$6

with

$t>0$7

For small $t>0$8, or near $t>0$9, this reduces to the linear law

$d = x(0)-x(-\tau)$0

with a slope $d = x(0)-x(-\tau)$1 that decays roughly as $d = x(0)-x(-\tau)$2 for large $d = x(0)-x(-\tau)$3. This reproduces both the observed linearity and the decrease of slope magnitude with $d = x(0)-x(-\tau)$4, while also identifying a regime in which linearity must fail as $d = x(0)-x(-\tau)$5 approaches its $d = x(0)-x(-\tau)$6-dependent minimum (Knotz et al., 8 Jul 2025).

5. Dependence on length and time scales, and the variance of back relaxation

The 2025 study emphasizes that MBR depends not only on the observation time $d = x(0)-x(-\tau)$7 and conditioning time $d = x(0)-x(-\tau)$8, but also on the cutoff length $d = x(0)-x(-\tau)$9, especially through statistical uncertainty. In living A549 cells, with fixed $b = x(t)-x(0).$0, $b = x(t)-x(0).$1 rises with $b = x(t)-x(0).$2, reaches a maximum, and approaches a plateau by $b = x(t)-x(0).$3. The long-time value $b = x(t)-x(0).$4 decreases strongly with $b = x(t)-x(0).$5, becomes negative around $b = x(t)-x(0).$6, and tends to $b = x(t)-x(0).$7 for large $b = x(t)-x(0).$8. Over a broader scan, $b = x(t)-x(0).$9 was varied from approximately $1/2$00 to $1/2$01. In passivated cells, where passive microrheology typically gives $1/2$02, the long-time MBR remains closer to $1/2$03 over a broad $1/2$04 range (Knotz et al., 8 Jul 2025).

The RHC model reproduces the same qualitative phenomenology when memory is present through coupling to $1/2$05. For

$1/2$06

MBR rises, peaks, and decays to a $1/2$07-dependent asymptote. As $1/2$08 increases,

$1/2$09

crosses from positive to negative and decays as $1/2$10 for $1/2$11: $1/2$12 There is also a critical activity,

$1/2$13

at which the large-$1/2$14 approach changes sign. In the small-$1/2$15 limit,

$1/2$16

so the long-time value is controlled purely by the activity ratio (Knotz et al., 8 Jul 2025).

To quantify statistical uncertainty, the same work introduces the variance of back relaxation,

$1/2$17

For stationary Gaussian processes, MBR itself is $1/2$18-independent, but VBR depends on $1/2$19 through the conditioning. Writing

$1/2$20

and

$1/2$21

one finds

$1/2$22

with

$1/2$23

VBR diverges both as $1/2$24 and as $1/2$25, reflecting two distinct pathologies: near-zero denominators and vanishing numbers of accepted events. It therefore has a unique minimum at $1/2$26. In the limit $1/2$27, the minimum approaches

$1/2$28

For finite $1/2$29,

$1/2$30

with

$1/2$31

The practical rule of thumb is therefore to choose $1/2$32 of order $1/2$33 (Knotz et al., 8 Jul 2025).

6. Empirical estimation, non-Gaussianity, and limitations

For discrete time series, MBR is estimated by scanning all index triples consistent with the chosen $1/2$34 and $1/2$35, forming

$1/2$36

retaining only those with $1/2$37, and defining event-wise back relaxation

$1/2$38

With $1/2$39 accepted events,

$1/2$40

and the sample variance is

$1/2$41

A standard error estimate is

$1/2$42

Bootstrap resampling of the accepted $1/2$43 values can be used for confidence intervals, and Gaussian predictions based on the MSD can serve as benchmarks (Knotz et al., 8 Jul 2025).

Several parameter-selection constraints follow directly from the variance analysis and the experimental setup. A finite spatial noise floor, approximately $1/2$44 in the reported measurements, contaminates very small $1/2$45, so $1/2$46 should be comfortably above that floor and near the theoretical optimum, approximately $1/2$47–$1/2$48. Very small $1/2$49 close to the sampling interval can bias MBR and increase variance; in cellular data, $1/2$50–$1/2$51 was effective for correlating with $1/2$52. Large $1/2$53 reduces $1/2$54, whereas very small $1/2$55 admits ill-conditioned ratios. Likewise, $1/2$56 should be large enough to reach the MBR plateau, but not so large that VBR becomes unnecessarily high (Knotz et al., 8 Jul 2025).

The comparison between data and Gaussian theory reveals a systematic limitation of purely Gaussian descriptions. In A549 cells, the measured $1/2$57 has the same qualitative divergence pattern as the Gaussian prediction, but its magnitude, especially at larger $1/2$58, exceeds the MSD-based prediction. This indicates non-Gaussian statistics in the underlying intracellular dynamics. The same conclusion is supported by the mismatch between the RHC prediction

$1/2$59

at low frequency and the robust cellular scaling

$1/2$60

A plausible implication is that intermittent active bursts, heterogeneous viscoelasticity, or non-linear couplings contribute materially to the observed trajectories (Knotz et al., 8 Jul 2025).

The main interpretive limitations follow from the assumptions behind the exact results. The criterion $1/2$61 requires stationarity in confinement, a finite mean position, and ergodic relaxation; in cells these requirements may fail. The Gaussian closed forms for MBR and VBR apply to stationary Gaussian processes only, so observed $1/2$62-dependence of MBR or excess VBR should be interpreted as deviations from that benchmark rather than as contradictions. Finally, the linear relation between long-time MBR and $1/2$63 is phenomenological, albeit experimentally robust over the stated $1/2$64-range, and the RHC model makes explicit that linearity must eventually break in some parameter regimes (Knotz et al., 2023, Knotz et al., 8 Jul 2025).